The Feature Extraction And Targets Recognition Based On Vector Hydrophone

The feature extraction of low frequency line spectrum from the radiated noise of surface and underwater targets is one of the key techniques in signal detection and target recognition. It is a very important topic of underwater acoustic signal processing, because it has direct influence on the effects of signal detection and target recognition. In recent years, the signal processing methods based on vector hydrophone have been presented and shown the benefits better than pressure signal processing methods, and therefore become hot topics. And much work has been done. The methods of vector signal processing have provided a new way for the feature extraction of targets, the detection and location of weak signals and targets recognition.In this thesis, the theory, algorithm and applications combined with the project "research on the feature extraction and analysis of XX acoustic vector" are discussed, according to the present development, main problems and practical needs in application of feature extraction of pressure and acoustic vector signals. The research of this paper is based on the single vector hydrophone. The feature extraction and recognition of surface and underwater targets are realized using the methods of vector signal processing and the principle of second and higher order statistics, combined with LOFAR, DEMON, nonintegral dimension spectrum and cross-bispectrum. Therefore the application areas of vector hydrophone are extended.The main research contents of this thesis are as follows:(1) The line spectral feature of radiated noise received by vector hydrophone are studied based on the theory of higher-order statistics, according to the nonstationarity of radiated noise of targets.(2) The estimating algorithms of bispectrum and cross-bispectrum of acoustic vector signal are proposed by the use of the algorithm of bispectrum and the definition of cross-bispectrum in scalar signals and are applied to the feature extraction of acoustic vector signal. The abilities in enhancing basic frequency signal and suppressing Gaussian noise are also discussed.(3) The feature extraction methods for nonintegral dimension DEMON spectrum in pressure and acoustic vector signal are discussed. The peak-to-background ratio (PBR) and probability of detection (PD) of propeller shaft frequency about nonintegral dimension spectrum of pressure and acoustic vector signal are discussed under the various noise background and the different SNR.(4) The target recognition techniques based on vector hydrophone and LMBPNN are discussed, using the composed features in acoustic vector signals obtained above, and the similar techniques based on vector hydrophone and RBFNN are also discussed, then the results are analyzed.(5) Simulations and real data processing are given, and the results testified the effectiveness of the above algorithms.This paper aims to get enough knowledge of line spectrum in acoustic vector signal and extract as more features as possible so as to improve the recognition-rate of targets.